long vacuum full, gin index and unusually long delete

It does look that on the box I had the gin index problem I'm having
other "strange" behaviours...
VACUUM FULL may take too long and suddenly a delete that took
around 2 min now seems it is never ending even if I just vacuumed
full and restarted the server.

I'm wondering if too high maintenance_work_mem or other cfg may end
up in such problems or a) there is a bug somewhere b) the box is
broken[1] c) some package is broken or can't coexist with some other
d) ?

I've experienced piping scripts through psql, see it die (or at
least disappear from the list of active processes without any log)
and continue to see postgres process eating 100% CPU.

need some advice on how to track down the problem...

This is an etch box with postgresql backported.

[1] even if postgresql is the real weight lifter there, so it is the
one that easier may stumble in some HW problems... for all the other
activities the box seems to behave happily... including some other
heavy weight activities where postgresql is involved.

--
Ivan Sergio Borgonovo
http://www.webthatworks.it